Before I get too far, VPS stands for Virtual Private Server. It’s like a dedicated server (in that you have full software control) but you run on a portion of the actual server with limited resources meaning hosts can have several customers on one physical machine, making them much, much cheaper.
VPS.net is one of (literally) thousands of companies selling VPSes.
At first glance VPS.net offer a lot and it’s true. They do. Tons of “cloud”-style features that make most other VPS companies look fairly static: Good backups, high-availability setups, fast and flexible provisioning, an API for automated provisioning, scheduled improvements, US and UK locations, CDN integration, super-high-speed Fusion IO storage and importantly, good hardware and fair pricing.
A lot of the more advanced features require extra cash over the base rate but given that Fusion IO and high-availability setups would otherwise cost a lot of money, this seems fair.
But features aren’t everything. They each account for very little when things don’t work well or at all.
As somebody who hosts my clients’ and my own websites, I deal with a lot of hosting demands. I have three virtual servers with three different companies but I’m only comparing two here: Linode and VPS.net. These accounts are near-equal in specs. My Tagadab server is much cheaper, with less RAM and CPU.
Problem one: The hardware is slow
I spend quite a lot of time connected to my servers. Linode just feels a lot faster than VPS.net (quicker login, quicker aptitude, etc) so I decided to compile and run unixbench
(version 4.1-wht.2). I also tested my PC (an Intel i7 with an SSD) to see how these shared computers compare to known desktop hardware.
Host | Dhrystone | Whetstone | Execl | File copy (1024) | File read | Final score |
---|---|---|---|---|---|---|
Linode | 580.6 | 129.1 | 209.9 | 682.6 | 1460.9 | 240.3 |
VPS.net | 289.1 | 141.9 | 74.9 | 264.9 | 493.1 | 109.7 |
My PC | 1044.5 | 312.9 | 857.6 | 3001.2 | 3195.8 | 1080.2 |
The first three benchmarks are CPU-bound. The next two are disk bound and the last one is an average of the full set of 11 tests (that I’m not including here for brevity).
But the outcome is clear: VPS.net is way behind Linode in performance.
Problem two: Downtime
Downtime is basically time when you can’t use your server. Be that through network issues, storage problems or the hypervisor blowing up, the result is the same: you can’t run your websites and your clients complain.
With both Tagadab and Linode I currently have 170 days of software uptime. Before those restarts it was something similar again. The majority of a year without any hint of a problem. There have been a couple of DDoS network attacks and network maintainance that stopped people being able to access the sites but they have been short, well announced and out of hours.
VPS.net has been a different story. We had some downtime when we first joined, some downtime early in the year, some downtime over summer and some downtime last night (that prompted me to write this epic whinge).
I realise that bad things do happen and they’re often non-obvious and hard to fix but the dedication to fix these issues has to be assessed and in their case, VPS.net appear to want to add more servers and data centres than fix existing issues.
On a personal level, when a server blows up and it’s not my fault, I’m the one who pays for the downtime. I have to coddle my clients, find out the issues, report them to the host, get status updates, make sure everything works when it’s back online. It eats into my time and that eats into my profits. I shouldn’t have to worry about a server going down but when it comes to VPS.net, I do.
Problem three: Reporting & remuneration
VPS.net have a status blog where they post out any known issues… But what is the point of this if you can’t keep it up?
Imagine the scene: You’re in France. You’re having a lovely time and then you get an SMS from a client: “Oli, the server’s down!”. You go to check the status blog on your horrifically expensive roaming 3G connection and… Nothing. Their status blog is down.
Yeah. That required me to do all the leg work myself. I had to log into the main site (which uses heavy dollops of javascript and images), find that the server is actually down, try to reboot it, fail and send a support message. Rinse and repeat for reading the reply, only to find out that there’s nothing I can do save begging my client to be patient.
The whole incident cost me £30 in data costs but when I asked for some form of goodwill remuneration but instead I got:
I will take all of your fedback under serious consideration as we move forward to help us improve.
In my eyes, if you cock up to a customer, you make up for it. You don’t do this by patting them on the back and thanking them for spending those hours logging incident reports, hours looking after clients and £30 data trying to find the problem when your stupid Wordpress install kicks the bucket.
Problem n: In short…
VPS.net have serious problems with their infrastructure. All their SANs and all their “self-healing” mojo counts for nothing when neither survives a squirrel’s fart.
They’ve got a nicely designed website, a nice concept of stacking nodes together but an appalling implementation that performs less than half as well as an equally specced-out (but cheaper) Linode box.
Come the year-end (save some major improvement), I’ll be moving away from VPS.net and probably buying another Linode.
If you’re with VPS.net or are considering them, have a look at Linode instead (that’s an affiliate link). Beaches and margaritas here we come! But seriously… Their website might not look as fancy but they offer a far better service.